This page has some details on doing political trend modeling.

Much of this work was done in the 2014 time frame. The created document contained topic modeling results from states that have enough headlines about their senators and representatives to do a topic modeling analysis.

The generation of the document and chart images, etc., were automated so that they could be done at a server for dynamic publishing via web site, email, etc.

1. Update information on current Senators and Representatives.

The web site used to do this at the time later stopped maintaining that information (after the project had completed). 2. Collect headlines (via XML feeds, etc.) for a period of time. Some of these were local news to a specific geographic location. Other headlines were national (or global). 3. Select all headlines (and paragraph) which contained a geographic mention and a Senator or Representative mention. 4. Perform a topic model analysis of the selected headlines for each group and for weekly time periods.

Cluster computing was used to do the computations. 5. Use the topic modeling results to create presentation graphics of the topic model results over time.